Sadly based on the data sheet I got some MPSH10s and as reported they didn't perform as expected.
The devices I found worked better for me are:
2n2222A(to18), I tried PN2222 and they produced less power (better than 3904).
2n2219A, the same as 2n2222A (they are the same die in different package)
I also tried a pair of irf510 as drivers with suitable mods for bias and they worked
fairly well. The transformers for predriver and driver were less than optimum using
those. I also tried the 2n6661As and they were about the same as IRF510 and
for those that might consider it 30* the cost of IRF510, and 9* the cost of 2n2222A.
Others that might work if you have or can find them are:
MRF495, 2sC799, 2sc1306, 2sc1307, 2sc2166
Most of the effort to get not more power as you will get that but more even power
ended up with changing the transformers for better impedance matching at higher
frequencies. My goal was 10W plus or minus a few and decent power at 10M.
If your pushing for more power in the 80-17M range the 2n2222A is a winner
in the driver and pre driver along with bias changes already discussed.