Looking at yesterday's run (day 17) you can see that for the last stage of the run, he was walking constantly from approx. mile 32 until the end, and his HR during this walk was 110-125 for the most part excluding breaks. Note that a big chunk of this was walking downhill too. Contrast that to day 10, when he ran+walked uphill for 6 hours with a range of 105-120 mostly. I was doubtful at the first but after seeing this there is no doubt something fishy is going on.
I do have some possible theories as to what could be happening behind the scenes here that I've been thinking about:
1) Assuming the data he puts on Strava is genuine (not manufactured or edited) he could be getting separate sources for heart rate and GPS data. For example, he could be running with some type of chest strap which measures heart rate and cadence for him, while someone else (or the car) has the watch and records the GPS track. This would open many possibilities for him: e.g. riding a bike, treadmill inside the van, etc. (Perhaps this is what is being done, and out of fear from accusations he decided to "run" day 17 with clean data and no cheating)
2) Also assuming data is accurate, he could simply be passing around the watch to other runners and "taking turns". However, this is unlikely since the data from runs where he was accompanied by someone else is also not 100% clean, and he wouldn't be able to pull this off with them seeing this (unless they've been paid to not tell - but this makes no sense on his end)
3) Then there is also the possibility that everything that is on Strava is completely manufactured, and is simply the result of overlaying the GPS tracks that the team did throughout the day, with cadence and heart rate measurements. Again for this, I also find it quite hard to believe, since it's a lot of effort and besides, if they are doing this, why not just post "believable" HR data?
All in all, I think something weird is going on, but the evidence against him needs to be more concrete. It's not robust enough to blame him for inconsistent HR data, so there needs to be a more systematic way to prove the inaccuracies and unclean data. It may well have to be video evidence after all that will do it, like in the previous cases we've seen commented here.