The 68000 was pretty predictable and easy to write assembly for. Even up to the 68030 Motorola was still valiantly trying to document the cycle counts of instructions, but the tables were getting pretty big and complicated and the customers pretty much stopped caring (collapsing hundreds of cases into "Okay, indexing is slower, we get it"). The compiler writers I sat next to never relied on the timings anyway, they just ran code. Against various chip revisions if they were smart.

ARM assembly was also pretty pleasant to work with, and earlier ARM CPUs were a lot more like 6502s than the modern plumber's nirvanas of speculation, multiple cache levels and cache coherency protocols.