By now I think parsing is not exactly the bottleneck. But back in 90s, I'm sure that it has been a part of why TP and BP were as fast as they were, esp. compared to C++. Having precompiled units (and therefore no need to #include hundreds of kilobytes of headers for every translation unit) is probably a bigger deal today.

Also, IIRC Delphi had only recently gotten generics, and they're still not heavily used? I suspect that code that's heavy on those will compile pretty slow.