The lack of reproducibility of mouse studies in humans in something that has been complained about in the comments here, but a recent paper in PNAS really drives the point home.
Prion diseases are fatal, untreatable neurodegenerative disorders caused by misfolded proteins which accumulate in the nervous system. What makes them especially crazy is that they can propogate by inducing normal proteins to misfold and clump together, causing the build up of damaging plaques.
Quinacrine is an approved anti-malarial drug that has been in use in humans since the 1930s. In the early 2000s, several studies screening for anti-prion drugs in mouse cells found quinacrine to be an effective treatment. Cells treated with the drug were able to clear the misfolded protein, and delayed the disease in mice.
The PNAS study decided to test quinacrine in their prion disease model, which uses deer and elk cells and protein. (Note: These are species that experience prion disease in the wild.) They found that quinacrine actually increased the spread of misfolded, disease-causing prions. This is completely opposite of the results seen in the mouse studies, and (surprise!) recent clinical studies of quinacrine in patients have found no beneficial effects.
Studies in mice are a powerful tool in narrowing down the plethora of potential treatments out there, but unfortunately it isn't enough. Testing in other species and human culture systems can help further narrow down targets, but they come with their own ethical quandaries. Hopefully our understanding of the human body and technology will develop to the point where we aren't so blind in our drug development. In the meantime, a reminder to take animal study results with a grain of salt. If they were all true, we would've cured cancer a thousand times over at this point.