Currently using the Pico 2 with an RP2350 chip. I'll try that branch and report back. Thanks.

I think I played around with shift registers using ESP32 and i2s. There were two things I ran into.
First, understand that the pulse timing is critical and doesn't tolerate too much variability.
The shifter registers needed latch that was timed slightly after the last bit of parallel data was pushed in. This was problematic as a whole channel had to be used for the latch.
The pushing of bits has to be fast enough, so eight (shifter register size) bits are sent then latch for each "bit" of stream data.

On RP the actual PIO program could manage this probably pretty well. But it all comes down to throughput if it's really interesting. Can you receive and put into DMA memory fast enough to keep the output going? Is there even enough memory for a full "frame" of that many streams of that many LEDs?

Makes sense. Yes. The PIO can handle the latching data in and out of the shift register. Slightly modified .asm code from the plain ws2811 code can handle both the strobing as well as the pixel pulses.

.program neopio
.side_set 2

.wrap_target
    set x, 7            side 2
    pull

bitloop0:
    set pins, 1         side 0
    jmp x--, bitloop0   side 1
    set x, 7            side 2
bitloop1:
    out pins, 1         side 0
    jmp x--, bitloop1   side 1
    set x, 7            side 2
bitloop2:
    set pins, 0         side 0
    jmp x--, bitloop2   side 1
.wrap``

96 streams would be pie in the sky theoretical based on just sending ability of PIO. Probably 32 or 48 is more realistic.

#884 (reply in thread)
Without using external shifter register. But this uses a lot of pins of the RP2040/RP2350.

Strip works now (off by one on the pin :)