Here are my VERY broad patent that claims X and Y.
USPTO: You are denied because this prior art A discloses X and Y.
Applicant: OK, here is my amended claim: X, Y and Z!
USPTO: But prior art A combined with B shows X, Y and Z, and its obvious to combined A and B.
Applicant: No it's not obvious but still they don't show X, Y, Z, and W.
USPTO: Okay A and B don't show that here is your patent.
The issue is that when you get so far down the argument the examiner gets sort of tunnel vision. They never spend a lot of time to see if a new reference they had not yet found showed X, Y, Z, and W.
Worse, sometimes things like W are obscure and wouldn't normally be mention in prior art. The argument then becomes, "okay this DRM server was disclosed in prior art, BUT my DRM server includes a memory caching process to reduce loading times for common files."
If you aren't super familiar with servers, it might not be apparent that all servers have something like that, or if they don't, it's obvious. Even if you are, you'll almost never find something like that explicitly spelled out in a previous patent. So they get the patent.
This actually wouldn't be a problem if patent trolls didn't exist. It is more efficient to deep test a patent only after it is used. The vast bulk of all patents are never used.
Also, some patents are strong, and some are of low quality. The low quality patents exist, in part, because the patent office does not have the resources (i.e., enough personnel, training and search technology) to perform a proper analysis. Those resources are lacking because congress takes funds from the patent office (the patent office could be otherwise self funded if not for congress taking the funds to use for their pet projects).