There's probably not much real performance difference. Just know that almost nothing I tried worked because optimising Scratch projects is more complex than trying to make sense of the Sega Saturn.
The original was the fastest polygon filler I could find, but I still wanted to see what optimisation I could do first before adapting it to my SCRATTT2000 STNICCC2000 implementation. Here's the test versions I've released thus far (don't worry about the out-of-order IDs, I was reusing an old project or two here): * Test 2: https://scratch.mit.edu/projects/1234915372 * Test 1: https://scratch.mit.edu/projects/1240121891