I couldn't agree with you less Ed, and that's saying something since you're a practical fellow just like me. Suppose we have three pumps; two running and one idle. Now suppose we have an overstress [I mean overlaod] on the one idle pump, and we hand-wave it away as you suggest.

What, pray tell, should we do when the plant is running at reduced capacity and we have one pump running and two pumps idle? Note that the load on any given pump nozzle in a group of parallel connected pumps is governed by two, usually conflicting, phenomena. (1) The headers are expanding from their last horizontal restraints or anchors that, in general, create a horizontal load set where the sum of the loads on the pumps and on the last restraint are zero. (2) The differential expansion of the hot and cold drops to the pumps is creating localized loads on the nozzles that, in general, sum to zero.

If the drops to the pumps are sufficiently stiff, then (2) can dominate over (1). This will make it very, very difficult to cope with the run/stop cases. A good design, however, has the drops to the pumps much less stiff than the header, so that (1) dominates over (2). When you reach this state of nirvana, your run/stop cases become much easier to manage.

One practical fix for this for a three pump system is to have the piping for the last pump on the header to have an expansion loop in both the suction and discharge drops. [This drives process engineers crazy (another reason for doing it!) because the flow is no longer "balanced" among the group of pumps. Sure. Right. Every pump that comes off the assembly line has EXACTLY the same head-flow characteristic, and every spool of pipe and every fitting has the exact same bore diameter.] But if you make the connections to the third pump flexible enough, the header can now squirm enough to balance the loads on the first two pumps without messing up the loads on the third pump.

With more than three pumps, it's even easier since you can group them in rows of pairs and triplets, with a main header and sub-headers. Again, this will drive your process engineers crazy, but it's the right thing to do.