Anyone have an idea when that might be implemented?
USB host MIDI already works on Teensy 4.0. So does regular serial MIDI.
Sometime in October is pretty likely for USB MIDI device. It will appear here (this forum) first as a beta test version, so keep an eye on the announcement section if you want to get it as early as possible.
Regarding higher performance, the very best thing you can do would be to create a test case that can be run on a single Teensy without any other hardware, or perhaps 2 programs for 2 Teensy boards connected together. Try to build in some sort of measurement of how well it's keeping up. Maybe use micros() or elapsedMicroseconds variables and occasionally Serial.print() that info about timing? Or if that changes the timing too much, maybe toggle pins with digitalWriteFast(), which can be measured with an oscilloscope or logic analyzer to check how well it's performing.
If you create something like that and I'm able to run it without any other MIDI hardware or special software on the PC side (but connecting an oscilloscope is perfectly fine), I'll give it a try here if you remind me when the early betas are published. I can't make any specific promises in advance, but that sort of performance test being available while the code is still under active development is the very best way to get me to optimize it well for your needs.
On the USB support in general, over the last 4 weeks nearly all my engineering time has gone into speeding up the Arduino Serial Monitor. The sad reality is Teensy 4.0 with Serial.print() sends so fast that the Arduino IDE runs out of memory and locks up. That really had to be fixed. Fortunately, that work is pretty much completed and I wrote this article about the process.
https://www.pjrc.com/improving-arduino-serial-monitor-performance/
The next major step on USB device support is unfortunately not MIDI. We're still not fully USB standards compliant, mainly regarding bulk endpoint max packet size and the "other speed" descriptors. The code is also not yet well optimized, and of course I've been putting off more optimization work while the already less-than-optimal code could crash the Arduino software. Now that the serial monitor can handle sustained high speed data, I can start working on the Teensy side.
USB device MIDI and the other USB types are a high priority. They will be implemented. Unfortunately this stuff all takes time, almost always longer than originally anticipated, and a number of pretty important things need to happen to have a solid foundation to build all that USB functionality.
Getting the USB code well optimized for 480 Mbit/sec speed is pretty likely to bring you the much higher MIDI performance you're seeking. Today we're achieving only about 15% to 20% of the theoretical speed. I am going to get that up much higher. Even though it means delaying MIDI and other USB types a little longer, focusing on that first while the code is still simple (only USB serial) is the best path to ultimately achieving excellent performance for MIDI, Audio and all the other types.