One of the disadvantages of using SaaS for regression testing is that you need to allow outside organization access to your pre-production test environment. In most organizations, it has significant part of real production information, which can be security risk.
Even if you de-identify user data (scramble addresses, phones, names etc) in pre-production testing (which might be significant and repeated effort), industrious industrial spy agent in SaaS company can ferret out significant knowledge about your business. Hence the dislike for it.
Some SaaS companies provide also "on-premises" testing cloud, which avoids such concerns (for added price of course). Or, companies just build the in-house test cloud themselves, if they have the expertise.
In-house solution using open-source might seem less expensive, but only if you do not factor in the cost of your own in-house experts learning the tools and managing the solution. Usually, buying something mass-produced by experts is cheaper than building it in-house from scratch (if mass-produced version can reasonably cover your needs).
Best of both worlds seems to be to buy expert "on-premises" solution build on open-source software, to avoid vendor lock-in, and learn from experts how to run it.
Cost (and feasibility of buy/hire vs sourcing in-house) also depends on the job market and cost of living: it might be more expensive (per hour) to hire admin in USA than in a developing country. This way, open source/cloud computing can shift opportunities to countries with lover cost of living. So it is kind of funny that many of us spend our own free time to teach others (for free) to do what we do for living in countries with high living costs...