I dislike the term RIP (unwisely termed RNA Immuno-precipitation--because, one does not precipitate RNA using an anti-RNA antibody), but that is not your fault and I am not blaming you--I am blaming the half-literate person who first used this term.

That aside, and sorry about that; but can you use a region of the 28S rDNA for positive control RTqPCR?. Do you need to care whether GAPDH or ACTB RNA is effectively precipitated? You should go for a positive control (rDNA), a negative (I hope you can mine many expressing lncRNAs for HEK293--there have been many recent reports on lncRNA transcriptomes, but I cannot remember if they included HEK293; you can also mine an eRNA as a negative) and your experimental targets. Of course, there always lies the dangerous possibility that a ncRNA could still be translated (and thus spook your analysis). Sorry, that's biology.

However, RNAs are notorious for binding beads (yes; I have been pained personally). And, moreover, many robosomal proteins are part of the CRAPome (http://www.nature.com/nmeth/journal/v10/n8/full/nmeth.2557.html?WT.ec_id=NMETH-201308) and they pose serious problems for downstream analyses of results. So, brace yourself for a little frustration. All I can say, is: Good Luck.

Nefeli, I think that GADPH and b-actin are probably good controls, but it is your negative control (comparison) that that could be your problem. I don't think that the supernatant is necessarily a good negative control. Might you perhaps doing a simultaneous IP with beads coated with an innocuous antibody like Protein G? 

RNAs sticking to beads can be an issue. In the past when I performed IPs, I pre-blocked the beads with a solution of yeast tRNA, BSA, and glycogen. I then washed the beads and added my extract. You could try this and see if it helps with the control RNA levels.