Earlier quoted context omitted.
The problem is that the normal implementation of a breakpoint/step is replace a single byte with a 0xCC opcode (INT 3), which will be hit by every thread. Making it lock onto a single thread would require a conditional breakpoint, checking the thread ID every time that code path is executed, which might potentially use a lot of CPU. I'm not sure that kind of conditional breakpoint has hardware support. It's definitel…
Should be doable with help from the operating system to have thread-local storage for the code segment when debugging.
That sounds like a pretty good idea, would be nice to see gdb implement something like that.