I hear over and over that Suunto's RGBM is some proprietary undocumented algorithm. But it's really not. There are multiple papers (and a book) that describe exactly what the algorithm does. It is not a bubble model by the way, and it doesn't generate deep stops (unless you enable the deep "Pyle" safety stops). The particular version known as "Suunto RGBM", "Mares RGBM" etc. is a normal M-value based model, except that it uses three functions to reduce the permissible M-values for repetitive and reverse profile diving consistent with the "real" RGBM model.
http://www.scuba-doc.com/rgbm.pdf is one of the papers that contains the formulas (see the "RGBM/ZHL (Critical Parameter) SYNTHESIS" chapter, page 16 and following). The formulas are the f_rp, f_dp and f_dy defined on page 18.
This is the part that a lot of people criticize as being too conservative. In reality, if you understand how these functions behave, you can work around it. The key is to avoid repetitive dives within a certain surface interval window. Either do the next dive very quickly (though you might still have too much tissue saturation in that case), or have a good hour of surface interval before you go on the next one. There are plots of this function (f_rp) available on the Internet, or you can do it yourself in WolframAlpha. You will see that there's a time window where this variant of the RGBM will be very conservative for repetitive dives, but once you're outside that window it's really not too bad.
Also keep in mind that these functions are not just made up to make the computer more conservative. They are supposedly based on actual measurements (though I haven't checked the data myself), in combination with the (full) RGBM model. Now the model might be wrong, and the conservatism is really unnecessary, but I think it's more likely that there is at least some truth to this particular prediction of the model, in which case you will really want the conservatism (we're not talking about deco diving here, so these predictions are pretty separate from what was tested in the NEDU deep stop study which questioned certain other predictions of the RGBM). Diving a more liberal computer doesn't actually make your dives safer! It only hides the actual risk.
In addition to this, Suunto RGBM also adds mandatory safety stops when violating the 33ft/min ascent rate. That particular part is not openly documented (as far as I know), but it's not doing anything crazy in my experience as long as you have your ascent speed under control (unless something goes very wrong, your "mandatory" safety stop will never be longer than the regular 3 minute safety stop).
Now the Suunto EON is a different beast. It is one of the very very few computers available today that implement the actual RGBM bubble model (Suunto calls it Fused RGBM). The EON does not use the formulas that I talked about above. In recreational diving, it will behave in the same way, but if you go into deep or decompression diving, it will behave more similar to a VPM-B computer and generate "real" deepstops. Now publications for the "real" RGBM are a mess. The basics of the model are available, but - as someone who has recently tried to implement it myself - the details are often missing. I do believe Wienke when he says that the model has been checked against their dive databases, but if you do very deep or aggressive decompression diving, I wouldn't be surprised if the model or its implementation had some significant flaws.
In the end, for recreational diving, I think the algorithm doesn't really matter all that much. I personally think that the RGBM assumptions about bubble seeds across repetitive dives are plausible, and I would rather go with a computer that opts to be more conservative here. But I'm not sure if there's much support from actual empirical data on this, so I wouldn't want to make any claims that this is the one correct choice. In the end I tend to agree that for recreational use, the form factor, clarity of the display, reliability/quality and ease of use are are probably more important than the algorithm.
Edit: Found a plot of the reduction functions: They are on page 8 of this PDF by Suunto:
http://ns.suunto.com/pdf/Suunto_Dive_Fused_RGBM_brochure_EN.pdf
Actually the one hour of surface interval that I mentioned is still a bit short, but that's roughly the point at which the additional conservatism of this algorithm starts decreasing again. After two hours, everything should be close to a regular non-RGBM computer again.
