Yeah, it definitely deserves an EP! However my guess is that the lack of speed could be the issue here. The actual output only covers little compared to the input image size (because of the masking), but the filter renders with speed similar to a full coverage texture.
I took a look at the internals, and here is what I think could be done. Proper threshold clipping should be able to shave off a great deal of the rendering overhead. Moreover we have two high pass filters which probably are the main culprits here, and these will not go well in hand with the threshold clipping due to internal temporary tile rasterization (bitmap-based components need to render complete internal tiles, so clipping will not help much). The high pass comps serves to add depth to the height map (and shade the surface), but this should be possible to achieve via profiling alone.