For a quick test if the PWM-is capable of outputing the Bitstream fast enough I wrote a small testprogram ( by modifiying the rpitx-Source). This testprogram initializes the PWM-Periperal with 125 MHz Clock (125Mhz clock and 32 oversampling translates to 3,9 MSPS)

The testprogram generates alternating pattern of 0xFFFF0050 and 0xFF000020 - the
result looks quite workable (see picture below)

there are some glitches from time to time - which can be expected on an
multitasking OS - but maybe with DMA this issue could be solved aswell.