Yang et al (2013): "To assess whether a marker transgene could be inserted into an endogenous locus, we coinjected Cas9 mRNA, sgRNA, and a double-stranded donor vector that was designed to fuse a p2AmCherry reporter with the last codon of the Nanog gene (Figure 2A). A circular donor vector was used to minimize random integrations. To assess toxicity and to optimize the concentration of donor DNA, we microinjected different amounts of Nanog-2A-mCherry vector. Injection with a high concentration of donor DNA (500 ng/ml) yielded mCherry-positive embryos with high efficiency, with most blastocysts being retarded, whereas injection with a lower donor DNA concentration (10 ng/ml) yielded mostly healthy blastocysts, most of which were mCherry-negative. When 200 ng/ml donor DNA was used, 75% (936/1,262) of the injected zygotes developed to blastocysts, 9% (86/936) of which were mCherry-positive (Figure 2C; Table S1)."

So efficiency is inversely proportional to toxicity, they treated many more than 5-25 cells (the selection would occur at the level of the embryo), and there was only 1-10% rate of mutation detection. Also, they used "Superovulated female B6D2F1 mice". This procedure leads to chromosomal abnormalities and probably genetic instability so we would expect elevated presence of mutations at any given site: http://jhered.oxfordjournals.org/content/77/1/39.full.pdf

I'll have to look closer at the HDR aspect though (primers used, etc). But what may be going on that usually they detect the insertion via PCR: there is one primer to a sequence unique to the cassette and another upstream or downstream that should only be in the cells. Then the segment spanning the junction is amplified which supposedly is conclusive evidence of insertion at the correct location. The problem is you can get single primer amplification and also the homology arms required for HDR are likely to contain similar sequences to the "cell-only" primer. Eg: http://link.springer.com/protocol/10.1385%2F0-89603-258-2%3A...

Hwang et al 2013: "On the next day, injected embryos were inspected under stereoscope and were classified as dead, deformed or normal phenotypes. Only embryos that developed normally were assayed for target site mutations"

They don't seem to tell us how many embryos were injected. And that study does not appear to use any type of control group at all. AFAICT, that is exactly the type of study that is consistent with a selection effect.