I'll tell you my opinion, and how I implement it.

The only times I've ever operated PSK split were when I was (semi-)rare DX and I had lots of stations calling. (I was operating from a large station in KH2 back in 2012 and 2014.) Most of the time I only had one or two replies (at most) for each CQ, so it was easy to pick one out to reply to even on my frequency. Occasionally there would be so many that I couldn't even pick out a callsign fragment, so I'd go split.

I've never had to do that in any other situation, even during contests. Occasionally (like during the PSKfest last weekend) I could pick out two callsigns calling at the same time, and see enough to recognize each. I'd simply reply "two stations calling. First N7WE de W3HF..." and then at the end of that QSO send "73 N7WE de W3HF sk now KD9PCN de W3HF..." 

As to how to implement it, I find the best way is to use the software to split the Tx and Rx audio frequencies (on the waterfall), and leave the radio on a single frequency. Most software that I've seen lets you lock your Tx frequency, but move the receive around. (I use DigiPan, but I'm sure someone can tell you how to do it in fldigi.) Since you can see 2-3 kHz on the waterfall when you only need less than 100 Hz, it's easy to pick them out. My Guam CQs were something like "CQ de W3HF/KH2 up 100-300." Then with my Tx freq locked, I'd move around in the range 100-300 Hz high and find someone to reply to.

73,
Steve

W3HF