Yes and no.
What you describe would be useful, but the scope of its usefulness is likely far less than you might think. Basically, you'd only be capturing some long term effects that have very high effect sizes / strong correlations. The sheer complexity, variability, and size of the dataset means that you won't be able to pull out subtle interactions, even with a large number of individuals. Barring fundamental advances in GWAS methods, you're looking at huge expense for very little benefit.
Now, collecting and organizing disparate datasets that are already being generated is worthwhile. Comparison is difficult due to the different methods used, but a good investigator will be able to do a lot of in silico work to, at least, do some preliminary work on hypotheses. On a personal note, this is actually a lot of fun because you can often come across from very exciting hints to follow up on, all with a quick feedback cycle conducive to 'flow'.
Ultimately, and this gets back to science fundamentals, this would amount to a big fishing expedition. Good science will always be strongly hypothesis driven. Good bioinformatics is built upon good datasets, and that only comes from very careful hypothesis consideration, solid methods, and a good understanding of the biology involved.