A controversial argument:
Scrum was developed for:
1) Small teams of talented people.
2) Strong buy in from the client(s) who are effectively members of the team.
3) To compensate for the evolutionary accumulation of knowledge as development progresses.
1) It is less contractually defensible without extraordinary change control – the causal relationship between goal, budget and what is accomplished is often open to greater risk of litigation or loss.
2) Clients are often comprised of different factions attempting to undermine each other, and conflicts not resolved in contracts are often impossible to avoid, leaving the agency exposed to failure, caused by the client discord.
3) There are a lot of people in the industry who lack the discipline to work in this manner, and the addition of contractors often exacerbates the problem.
4) It used to be more difficult for agencies to attract top technical talent. This is declining but is still, to some degree, true.
Is that it is better suited to teams who work together all the time, and in particular for product development, and less comforting to use in high risk environments with a significant amount of customer management.
Given the tendency of the major agencies to have less trusting delivery relationships with their clients I would have to approach any question extremely cautiously less one or two major failures a year remove all perceived benefit from the broader financial and relationship questions.