its hard to answer the first question , yes the cpu may well get called away but its generally not for long enough to be an issue. There are many factors , depending on how io or cpu bound your application is that would decide if this is a problem. i would suggest you do some exploratory testing on a small scale to find out the practical limits of how much delay it can cope with.
you could get a reasonable approximation by running your applications and target oses in guests on a local hyperv host and compare against non virtualised versions of the same setup. i would suggest applying some load in another guest VM such as compiling a kernel or using fio to keep the disks busy. make sure that host has a cpu that properly supports virtualisation to keep it fair.
for the second question , no it wont be predictable but if it happens it will generally happen to everything around it so ensure your mitigation doesn’t make it worse for something else.