I am not sure that's how it works. Are all RH orders not-held? That would be possible but surprising given the regulatory scrutiny. It is even more unlikely to be 1-30 seconds since Rule 661 requires execution at NBBO or better within a second, IIRC. This is what a PFOF model looks like as far as I know (see disclaimer below). A customer sends a market or a limit order. Citadel or KCG gets the first look (from my understanding, the flow buyer pays for that optionality regardless of the outcome) and decides if it fits either their book or their alpha model. Some of resulting orders are outright gaming the order book (e.g. a customer can get an improvement on a small order, while they would now gain queue priority thats probabilistically more valuable) and some of it is simply "oh, yeah let's fill this guy cause we know the touch is going to shift any moment now". If it's a limit away, it's valuable from information perspective since you know that if you are done and don't like it, you can fill the customer order at a negligible loss. Disclaimer: Thankfully, I do not deal with PFOF, so everything I know I learned by osmosis from my HFT colleagues.
Let’s say you are actually trading 50k shares. In stocks that are liquid and low dollars I have never had this issue (when I trade 50-100k shares through my podunk PFOF happy broker). In illiquid stocks I have, but it’s 10 cents. But that cost is embedded into the risk premium.
Well I pay a fraction of what the dma brokers charge which (as said before) is a real cost savings. It’s unclear how much that option I’m selling is worth (a soft cost).
He does get lower commissions, thats an indisputable fact. He might or might not increase his total cost of trading. FWIW, I've only heard of one retail broker that explicitly does not get involved in PFOF and that's IB. In fact, that's there current selling pitch "IB does not sell its order flow". Predictably, that's because they're giving access to their ATS to liquidity providers (i.e. HFTs) and now get paid "commissions or commission equivalents". Optically it's better but IRL it's the same thing.
Do you do TCA(Transaction Cost Analysis) on your executions? If you're unclear on the cost of your broker's routing procedures, why not quantify it?