Like I said I know most companies already fail at the basics. But these are normally well known already, just not fixed due to political pressure. Having the security team's management be better at influencing would pre-empt these issues. Pentests serve to bring these known issues under discussion again, but they often don't bring any unknown issues to light.
Also, they tend to be too artificial. In one example: In a recent pentest I know of, they sent some remote access malware to the admins, which they all ignored and reported. However as they couldn't continue they called one admin's manager and asked him to tell the admin to click on the link so that they could proceed. The outcome of the pentest was that if a malicious actor sent a malicious remote access malware and if it was clicked on, they would have remote access. Well duh.
Of course this could be mitigated by having separate workstations for email and admin activity, which was an issue that was already understood and a mitigation in progress. This is what I mean by pentesting not raising new issues. Another example: The company in question already test new web apps for known attacks, and is quite successful (in fact I've never seen a new app come through the certification process in the first round). What I'd expect from a pentest is to tell us something we don't know :) It's an interesting second-opinion but it's not the amazing X-ray it's promised to be.
> What you're suggesting is everyone has to be willing and able to write fresh exploitation on the spot when they're testing a client's network
Which is exactly what a serious adversary would be doing in a targeted hack!