BJT capacitance is smaller, and gm is much higher.
2N3904 is rated 200mA, and what, 8pF E-B? 2N7000 is rated 200mA, and is more like 30pF (at low voltages). BJT needs about 0.3V swing to go from reasonably-on to reasonably-off, while the MOSFET needs a volt or so. Or in practical switching circuits, about 1V vs. >3V.
I guess what's interesting is MOSFETs are a better self-match: even given that, in Si, PMOS performs ~2.5x worse than NMOS, the drain voltage and current range is very suitable for driving more gates. Hence the dominance of CMOS.
BJTs go well with themselves at low voltages (I2L is probably the best example), but aren't as easy to use at conventional logic voltages. Consider TTL's relatively high power consumption, and poor V_OH vs. I_OH. The latter problem could of course be remedied with PNPs if they were available in ICs at the time, but the voltage swing is still way more than needed for Vbe, while the current requirement is the same (set by storage charge and Ccb). So the power dissipation is higher.
Does mean BJTs work very nicely in RF applications, where networks can transform collector and base circuit impedances. I suppose MOSFETs still are dominant here today; LDMOS is huge, while RF BJTs are... does anyone make them anymore? Replacements for old equipment, or are those all NOS?* But this wasn't always the case I think; more that, now that they've both been heavily optimized, BJTs hit a physical wall (recombination) while MOSFETs kept going. Advanced relatives are current tech though (SiGe HBTs).
*Checking, yep, still stocked; don't know if they're classic PNs or what, but the ceramic stud/flange microstrip package anyway. And still as expensive as ever.

Tim