Thank you to an anonymous contributor for bringing to attention a two part blog post by Stephen Blinick. Stephen's post details how the performance of rtl_tcp can be significantly improved by modifying to code to use a ring buffer instead of using semaphore based locking. If you weren't aware, rtl_tcp is a program that allows you to run your RTL-SDR remotely, and connect to it over a network connection.
The result is a tremendous performance improvement in rtl_tcp according to Stephen. Before the changes he noted that his Raspberry Pi 3B+ could only support a sample rate of 1.92 MSPS over WiFi, and even that had 1-2 seconds of lag. After the ring buffer changes his Pi 3B+ can handle the maximum sample rate of 3.2 MSPS with zero lag. On his Pi Zero W he can achieve a sample rate of 1.92 MSPS over WiFi with minimal lag, whereas before he could only achieve 0.92 MSPS with huge 5-10 second of lag.
The patch is available as a pull request over on the Osmocom GitHub.
Unfortunately this patch might not be included in the official upstreamed Osmocom drivers because Stephen submitted the patch as a pull request to the GitHub, and Osmocom only accept patches via their mailing list. If anyone reading this is familiar with the Osmocom patch submission requirements, we'd like to encourage you to help submit this patch for consideration.