@Ben Webster asked me to answer this, so I will. It's a tough one!
The problem here is that I don't feel that PhD programs are all on the same page here, and their differences from one another are not easily visible to the outside (e.g. two departments of similar quality may well differ on the issue). Here are two different philosophies for departments that recommend but do not require the math subject GRE:
We don't take the math subject GRE that seriously. A really good score is positive, but it is certainly not sufficient to get you admitted. We don't use poor scores to exclude candidates...but we admit that all else being equal, higher scores are better. A missing subject GRE score just means that we don't take this part of the application into account.
We do take the math subject GRE seriously, and a low score is a red flag for us. However, although we will almost never want to admit a student with a poor score, we understand that sometimes really strong applicants just don't take the test. If we get a fantastic applicant who didn't manage to take the subject test, we would like to have the right to admit them anyway. Making the subject exam strictly required would call that right into question, so we won't do that. But if you are trying to best compete with the heart of the applicant pool, we advise you to take the test and do well on it.
So you see the problem. The 48 percentile score makes it quite likely that the OP is not (or not yet) a latter-day Gauss. Thus for a department with the second philosophy, he should submit his score. For a department with the first philosophy, maybe not...but there's another wrinkle.
The OP characterizes his 48 percentile score as "miserable". That is a misperception beyond mere histrionics: I believe this is actually a decent score for a range of programs. (I think it would be fine for mine, but I am no longer on the graduate committee so can't speak so specifically.) Let's get real: the 48 percentile is on a test which is only taken by people who are seriously considering math grad school but the converse does not hold: a lot of people who are considering math grad school -- and even many that are admitted to a certain range and class of programs -- are not taking this test. Further, I have heard rather convincing rumors that many of the highest scores on this exam are attained by foreign students taking the exam under what we Americans would politely call nonstandard conditions. Apparently this is true to an extent that it does skew the results a bit, so 48 percentile is probably not below average for "reliably honest" scores.
So the situation is even more confusing: I'm not quite sure, but I believe that there are US math PhD programs which fall under the first philosophy, but for which nevertheless turning in a score of 48 percentile would help, rather than hurt, the OP in terms of his placement in the pack.
Let's bring this back to the poor OP. He has gone to all the trouble (and money) of taking the test. As a result there is information out there about the OP's math skills that most PhD programs would want to know. Whether it is in the OP's best interest for them to know it is highly unclear: it varies in a way that is a bit opaque to me, let alone the OP. Because his score is under 50% he thinks it is bad and thus doesn't submit it to certain schools. (And maybe he is right.) Isn't everyone but the Educational Testing Service losing out on this proposition? What a mess.
For the math faculty who are reading this: I hope you now understand why I contributed in my own department's changeover from recommending the exam to requiring it. I recommend that you require it too (!). When people are taking the test, not telling us their scores, and we understand why they are doing so, then we are not setting things up in the best way.
