BCD instructions are slower than emulating them with more traditional operations and were rarely used in 32-bit software (16-bit is a different story). Most people would recompile their software and never experience this as an issue.

The removal of the segment registers was the only one that really reduced capabilities - VMWare's original solution became impossible because of that. However they were replaced by hypervisor instruction sets which are much more capable. The proposed solution in this article would never have fixed this though - which again is why its true but misleading.