If you have sequenced from enough no of individuals, then you can do the statistical analysis for correlation between the phenotypic/genotypic variability along with elevation gradient. You can use SPSS software for this purpose. Moreover if you are handling a big data, you can also use Estimate S software for sampling effort.

It is better to do Maximum likelihood analysis using Garli or other suitable software if yo have, MrBayes analysis can also get high support branch support value. In Summary, the final result is to find well congruence result for different branch. I don't think the combine analysis is good analysis method for different gene conferring different evolutionary history. As far as I know, there are total different result can obtained if you using separate gene data set or using combined gene data.

Depends on your specific question, and how many individuals you have sequence data for.

I have phenotypic data for 43 individuals representing 11 elevations. but i selected only 15 for my genetic data so that they cover low,mid and high elevations. unfortunately PCR;s worked only for 8 of Cyt B1 and 13 for IRF 2. However my main focus is the varriation along the elevation. 

How much information you can get will depend a lot on the amount and structure of variation. Since this is mitochondrial DNA, what I would do is align independently the sequences for each markers, make sure you trim all nucleotides of lower quality (it is better to be conservative and loose potentially informative variants than to risk having your results driven by sequencing errors), and infer phylograms from this data. Use either Maximum Likelihood (phyML, RAxML, phylip) or Bayesian Inference (MrBayes), or both. Do this separately for CytB1 and IRF2. Compare the trees between both markers and both methods. If they are more or less congruent (=similar, maybe differing in one or two bipartitions at most), label each individual with its elevation (by hand, on a printout, use nice colors) and check individuals from different elevations group together.

If you indeed find that trees are congruent and individuals cluster by elevation, then I highly recommend you try to get more data: more individuals, more sequences, more markers. It will be hard to achieve statistical significance with such a low sample size, unless it is so strikingly obvious that you wouldn't need statistics anyway.

If you find that the trees are not congruent, or that individuals from different elevations are all mixed up, then I would try to go for a different marker. Even if this is the case, your data is likely to be insufficient to confidently discard genetic structuring by elevation.