diff --git a/src/hardware/hw_cache.c b/src/hardware/hw_cache.c
index 79fb6f4cb..4b9da2616 100644
--- a/src/hardware/hw_cache.c
+++ b/src/hardware/hw_cache.c
@@ -511,6 +511,7 @@ static void HWR_GenerateTexture(GLMapTexture_t *grtex, INT32 texnum, boolean noe
 	for (i = 0, patch = texture->patches; i < texture->patchcount; i++, patch++)
 	{
 		boolean dealloc = true;
+		boolean doremap = W_NeedPaletteRemapPwad(patch->wad, patch->lump, false);
 		size_t lumplength = W_LumpLengthPwad(patch->wad, patch->lump);
 		pdata = W_CacheLumpNumPwad(patch->wad, patch->lump, PU_CACHE);
 		realpatch = (softwarepatch_t *)pdata;
@@ -523,16 +524,15 @@ static void HWR_GenerateTexture(GLMapTexture_t *grtex, INT32 texnum, boolean noe
 #ifdef WALLFLATS
 		if (texture->type == TEXTURETYPE_FLAT)
 		{
-			if (W_NeedPaletteRemap())
+			if (doremap)
 				R_DoPaletteRemapFlat(pdata, lumplength);
 			realpatch = (softwarepatch_t *)Picture_Convert(PICFMT_FLAT, pdata, PICFMT_DOOMPATCH, 0, NULL, texture->width, texture->height, 0, 0, 0);
 		}
 		else
 #endif
 		{
-			(void)lumplength;
 			dealloc = false;
-			if (W_NeedPaletteRemap())
+			if (doremap)
 				R_DoPaletteRemapPatch(realpatch, lumplength);
 		}
 
@@ -844,6 +844,7 @@ static void HWR_CacheFlat(GLMipmap_t *grMipmap, lumpnum_t flatlumpnum)
 	UINT8 *flat;
 	size_t steppy;
 	size_t size, pflatsize;
+	boolean doremap;
 
 
 	// setup the texture info
@@ -880,13 +881,13 @@ static void HWR_CacheFlat(GLMipmap_t *grMipmap, lumpnum_t flatlumpnum)
 	grMipmap->width  = (UINT16)pflatsize;
 	grMipmap->height = (UINT16)pflatsize;
 
+	doremap = W_NeedPaletteRemap(flatlumpnum, true);
+
 	// the flat raw data needn't be converted with palettized textures
 	W_ReadLump(flatlumpnum, Z_Malloc(W_LumpLength(flatlumpnum),
 		PU_HWRCACHE, &grMipmap->data));
 
-	// palette remapping
-	// have to do this here since the lump isn't cached
-	if (wadfiles[WADFILENUM(flatlumpnum)]->compatmode)
+	if (doremap)
 		R_DoPaletteRemapFlat(grMipmap->data, size);
 
 	flat = grMipmap->data;
diff --git a/src/r_textures.c b/src/r_textures.c
index 3dc4fb77e..dfff08149 100644
--- a/src/r_textures.c
+++ b/src/r_textures.c
@@ -355,6 +355,7 @@ UINT8 *R_GenerateTexture(size_t texnum)
 	UINT16 wadnum;
 	lumpnum_t lumpnum;
 	size_t lumplength;
+	boolean doremap;
 
 	I_Assert(texnum <= (size_t)numtextures);
 	texture = textures[texnum];
@@ -384,6 +385,7 @@ UINT8 *R_GenerateTexture(size_t texnum)
 			return block;
 		}
 
+		doremap = W_NeedPaletteRemapPwad(wadnum, lumpnum, false);
 		pdata = W_CacheLumpNumPwad(wadnum, lumpnum, PU_LEVEL);
 		realpatch = (softwarepatch_t *)pdata;
 
@@ -396,7 +398,7 @@ UINT8 *R_GenerateTexture(size_t texnum)
 			goto multipatch;
 #endif
 
-		if (W_NeedPaletteRemap())
+		if (doremap)
 			R_DoPaletteRemapPatch(realpatch, lumplength);
 
 		// Check the patch for holes.
@@ -482,6 +484,7 @@ UINT8 *R_GenerateTexture(size_t texnum)
 
 		wadnum = patch->wad;
 		lumpnum = patch->lump;
+		doremap = W_NeedPaletteRemapPwad(wadnum, lumpnum, false);
 		pdata = W_CacheLumpNumPwad(wadnum, lumpnum, PU_LEVEL);
 		lumplength = W_LumpLengthPwad(wadnum, lumpnum);
 		realpatch = (softwarepatch_t *)pdata;
@@ -495,16 +498,15 @@ UINT8 *R_GenerateTexture(size_t texnum)
 #ifdef WALLFLATS
 		if (texture->type == TEXTURETYPE_FLAT)
 		{
-			if (W_NeedPaletteRemap())
+			if (doremap)
 				R_DoPaletteRemapFlat(pdata, lumplength);
 			realpatch = (softwarepatch_t *)Picture_Convert(PICFMT_FLAT, pdata, PICFMT_DOOMPATCH, 0, NULL, texture->width, texture->height, 0, 0, 0);
 		}
 		else
 #endif
 		{
-			(void)lumplength;
 			dealloc = false;
-			if (W_NeedPaletteRemap())
+			if (doremap)
 				R_DoPaletteRemapPatch(realpatch, lumplength);
 		}
 
@@ -935,8 +937,11 @@ UINT8 *R_GetBrightmapColumn(fixed_t tex, INT32 col)
 
 void *R_GetFlat(lumpnum_t flatlumpnum)
 {
+	// FIXME: checking this every time a flat is fetched for drawing is stupid...
+	// sadly they are not cached anywhere else so i have no choice
+	boolean doremap = W_NeedPaletteRemap(flatlumpnum, false);
 	void *flat = W_CacheLumpNum(flatlumpnum, PU_LEVEL);
-	if (W_NeedPaletteRemap())
+	if (doremap)
 		R_DoPaletteRemapFlat(flat, W_LumpLength(flatlumpnum));
 	return flat;
 }
diff --git a/src/w_wad.c b/src/w_wad.c
index 37cc1dea8..52b8a00f0 100644
--- a/src/w_wad.c
+++ b/src/w_wad.c
@@ -1681,16 +1681,19 @@ void zerr(int ret)
 }
 #endif
 
-// only do palette swapping ONCE, when caching the whole patch
-// dumb hack but it works!
-static boolean dopalswap = false;
-
-// returns true if the most recently loaded lump is freshly cached, AND comes from a compatmode file
-boolean W_NeedPaletteRemap(void)
+// returns true if the given lump is not cached, AND comes from a compatmode file
+boolean W_NeedPaletteRemapPwad(UINT16 wad, UINT16 lump, boolean ignorecache)
 {
-	boolean r = dopalswap;
-	dopalswap = false;
-	return r;
+	if (!wadfiles[wad]->lumpcache[lump] || ignorecache)
+	{
+		return wadfiles[wad]->compatmode;
+	}
+	return false;
+}
+
+boolean W_NeedPaletteRemap(lumpnum_t lump, boolean ignorecache)
+{
+	return W_NeedPaletteRemapPwad(WADFILENUM(lump), LUMPNUM(lump), ignorecache);
 }
 
 /** Reads bytes from the head of a lump.
@@ -1881,15 +1884,11 @@ void *W_CacheLumpNumPwad(UINT16 wad, UINT16 lump, INT32 tag)
 	lumpcache = wadfiles[wad]->lumpcache;
 	if (!lumpcache[lump])
 	{
-		dopalswap = wadfiles[wad]->compatmode;
 		void *ptr = Z_Malloc(W_LumpLengthPwad(wad, lump), tag, &lumpcache[lump]);
 		W_ReadLumpHeaderPwad(wad, lump, ptr, 0, 0);  // read the lump in full
 	}
 	else
-	{
-		dopalswap = false;
 		Z_ChangeTag(lumpcache[lump], tag);
-	}
 
 	return lumpcache[lump];
 }
diff --git a/src/w_wad.h b/src/w_wad.h
index 8a088ad3c..b518b7d39 100644
--- a/src/w_wad.h
+++ b/src/w_wad.h
@@ -205,7 +205,8 @@ void *W_CacheLumpNumForce(lumpnum_t lumpnum, INT32 tag);
 
 boolean W_IsLumpCached(lumpnum_t lump, void *ptr);
 boolean W_IsPatchCached(lumpnum_t lump, void *ptr);
-boolean W_NeedPaletteRemap(void);
+boolean W_NeedPaletteRemapPwad(UINT16 wad, UINT16 lump, boolean ignorecache);
+boolean W_NeedPaletteRemap(lumpnum_t lump, boolean ignorecache);
 
 void *W_CacheLumpName(const char *name, INT32 tag);
 void *W_CachePatchName(const char *name, INT32 tag);