They're out there:
https://www.ti.com/lit/ds/symlink/tps40200-q1.pdfPerformance is king. The difference is not small -- PMOS is about 2.5x worse than NMOS. More die area = bigger package and more cost (with knock-on costs from lower production quantities -- i.e., PMOS in general are less common, and cost more as a result). Poorer Qg*Rds(on) figure of merit = generally higher switching losses, and either higher conduction losses or higher drive losses (drive power is Qg * Vgs(on) * Fsw).
You need a driver anyway -- the gate has generous capacitance in either type, so a mere logic-level output is unsuitable for all but the smallest transistors -- and it's not much added die area (if integrated) or cost (if discrete*) to include the bootstrap driver for NMOS. So it's very practical to use NMOS in high or low side application.

* "RTL" level discrete that is, i.e., using ICs for basic functions. Low side, logic-input drivers are cheap and plentiful (TC4427, etc.), as are bootstrap drivers (IR2110, etc.). If you're doing something truly discrete (bare transistors), it's harder -- a high-voltage, noise-immune level shifter is certainly doable, but it will take more transistors than the alternative (compare
this (first page, top and left half) to
this, for example).
Tim