1) db is a dimensionless label. It refers to the logarithmic representation of a ratio between two values. The term +2.5db tells us that something is a bit louder than something else, but unless you tell us what it's louder than, the statement is meaningless. 65db sounds like a lot, and in terms of dbfs, VU, V, or u it is a lot. 65dbSPL is actually pretty quiet. Always label your db's if you mean to reference an absolute level!
2) Modern DAW software can handle quite a bit higher than 0bdfs internally. It doesn't actually clip until you render the it to a fixed-point audio file, or try to push it out your DAC. That is, you don't really
have to turn down the individual tracks to get the master peaking below 0dbfs (though it is best practice). You can probably just turn down the Master fader without worrying about impacting any post-fader sends or dynamic processing which might be working in the mix. Then remember for next time to shoot for lower overall levels.
3) If there is a problem it likely sits in the realm of "crest factor" - the difference between the absolute loudest peak and the average RMS level of the program. If you open up a typical unmastered mix and zoom out to see the whole waveform, you will almost always see two or three "layers" of peaks.
First there will be a fairly dense row clustered around the middle (-inf dbfs) line. This is the "meat" of your mix, and kind of represents the basic percieved loudness. (Don't flame me! I said "kind of", you have to squint a little)
Next you'll see a less dense, but fairly regular row of peaks that generally represent the attack of the main drum elements. I generally consider this to be the real "top" of the dynamic range of the mix.
Then there will be a handful, maybe a dozen or so, peaks that poke up a few db above that. I call these "aberrant peaks". They are usually caused by accidental coincidence of a number of elements in the mix. Several things just happen to all be push hard in the same direction at the same. You might think that it would be a desirable thing, but it's pretty damn tough to do on purpose with real performances of real instruments, and it tends to eat headroom disproportionately to any positive "impact" which they might contribute, and usually want to be controlled.
There will almost always be another row - consisting of just one or two little peaks which poke up yet further. These are essentially errors. They usually only last a sample or a few and may not be audible even if pushed beyond the clipping point. They may be extreme examples of the "aberrant peak"/accidental coincidence phenomenon. They might be actual glitches in the processing, clicks caused by poor edits, etc. Doesn't really matter. They set the absolute maximum peak level of the file. If you try to keep these below 0dbfs your overall mix will be unacceptably quiet. They need to be squashed hard.
Now, it's best to deal with dynamics issues at the track/bus level in mixing, but you will still likely see these kinds of things.
For a truly quick and dirty (and generally good enough for most things) "loud enough" master, I'll usually bring up the mix in SoundForge and use its Normalize function twice. The first pass is in RMS mode, with "compress to avoid distortion" on. I'll usually shoot for RMS between -15 and -12dbfs depending on the mix. Some really crazy stuff can go up to -10 or so, and really open or dynamic mixes can sometimes be a bit lower, but it all depends. This will almost always squash the highest peaks by a couplefew db so that they hit 0dbfs. The second pass is in Peak mode to bring it back down to -0.63dbfs - a "safety margin" to keep from clipping cheap consumer DACs and/or any mp3 conversion issues. This could be done using the Volume process also, but I have somewhat esoteric reasons to do it this way. This is also about the same thing as slamming into an L2 (Waves brand brickwall limiter) as mentioned above.
For a slightly more serious "master" I will run a few compression processes in series. First a fast and hard compressor or limited set just above the "aberrant peak" region to squash the "errors". Then either a second, somewhat less aggressive compressor set to just catch the tops of the drum peaks and bring the "aberrant" peaks (and squashed errors) down closer to the average peaks, or I'll set this second a little more aggressively, but with threshold set to only get the aberrations and a third relatively light comp to just reign in all of the remaking peaks.
If it doesn't sound loud enough after all of this, you need to go back and examine the mix and arrangement. Is there a lot of extreme low frequency information which might not even be audible, but is eating up your headroom? Are there other "spectrum issues" causing it to sound thin or weak even when cranked? Does it need more distortion and snare rattle so's you know it's loud?
Hope some of this helps.