I had to make around 50 "linker-scanning" mutagenized promoter
constructs involving PCR amplication of a 800 bp promoter fragment.
Only a couple of times did I detect infidelity due to Taq. Curious
about proofreading thermo-pols as you are, I tried Pfu and found that my
PCR reactions would work only after substantial troubleshooting (and
even then, not very well), so I returned to using Taq. You might want
to give Taq a try, unless the newer generations of proofreading pols