I think you might need to add the 'conv=fdatasync' option to your benchmark there in order to get real results - otherwise you could just be looking at results from the hosts cache:
Even then, the host could be lying and telling your VPS that a write was successful when it was not actually written to the host disk. Still in the hosts disk cache, which means great performance but if the host ever crashes, all data on your VPS disk will likely be lost.
I know people who have been royally screwed by exactly this - they give you an API, devs everywhere come up with ideas, do the hard work - and the platform picks the low hanging fruit. Not only do they get carte blanche to steal peoples ideas but the licenses often give them complete rights to steal your source code as well.