This in particular sounds very fake to me.
If they are delaying the regular orders, then they are either a) having drivers sit idle or b) freeing up resources for the priority orders.
In the (b) case this would just deliver the promised prioritization behavior, not evil and not what OP is claiming.
In the (a) case where they are actually having drivers sit idle, then they are reducing the throughput of their system significantly. Which might be fine for a quick A/B test on a subset of customers, but as framed this is basically a psy op to trick customers en masse into thinking the priority order is faster when it really isn’t. To have that effect, you would need to deploy this to all customers long enough for them to organically switch back and forth between priority and regular enough times to notice the difference. That seems like it couldn’t possibly be better than the much simpler option of just implementing the priority behavior and reducing its effect down to zero slowly over time (which would be evil but isn’t the claim).