unix: Enable -fno-crossjumping for fast build.

Confirmed that it improves perfomance of simple "for i in range(N): pass"
loop by 15% on Core2.

unix/Makefile

@@ -133,4 +133,4 @@ uninstall:
 # build synthetically fast interpreter for benchmarking
 fast:
 	@echo Make sure to run make -B
-	$(MAKE) COPT="-O2 -DNDEBUG" CFLAGS_EXTRA='-DMP_CONFIGFILE="<mpconfigport_fast.h>"' BUILD=build-fast
+	$(MAKE) COPT="-O2 -DNDEBUG -fno-crossjumping" CFLAGS_EXTRA='-DMP_CONFIGFILE="<mpconfigport_fast.h>"' BUILD=build-fast