Generally, starting out in RGB and converting to CMYK at the end of the process is a good thing. How are you converting to CMYK? Is it just a mode change, or are you converting to a specific profile based on the printer's recommendations or perhaps, some other criteria?
How far off are the first round of proofs? If you're using a CMYK profile that is perfect for the proofing system and subsequent press, then your first round of proofs should be very very close to being perfect. If you're making any significant change on the second proof, then it's very probable that your conversion is not right for that proofing system. This, of course, does not take into account subjective decisions that you make once you see the proof. That is, the proof could be a very close match to your monitor buy after seeing the proof you want to change something even though that's how it appeared on screen.
All too often, printers will give you the wrong information about their process or claim they comply to SWOP or GraCol standards when they really don't.
In the event of having little or no solid information regarding printing conditions, about all you can do is to separate using Photoshop defaults. Because that's what so many people use who don't know what they're doing, there are a huge number of images that printers get that have been converted that way, which forces them to be able to print them at least somewhat in the ballpark.
My goal is always to take that a step further whenever possible, including pre-emptive strikes to keep skies from going to purples or grass too yellow that are likely to look better in print than if I didn't do them. I also typically choose black generation schemes that help the image to print better on press, but will not show an improvement at the proofing stage, as the proof can never mimic certain aspects of the dynamics of the press.