The same limits apply in Experimental (on the 112.5-merge branch, at least). Generating maps at 400 height and 7 roughness, for example, I get a mountain range to rival the Himalayas and yet still double height slopes (straight ones, at least) are hardly anywhere to be seen (there are one or two dotted around).
Perhaps the answer is in having different map generators. Using heightmaps is one solution, but it is much less convenient than using an in-built generator, and it gives rise to far less variety. I see the point about the need to retain the Perlin generator for enlarging existing maps - but perhaps we could look into having different generation algorithms for new maps? Even Cities in Motion 2, I think, has about 4 different generation algorithms, of which Perlin is one (and is one of the less effective ones, I think).