Accenture revenue in 2016, $32.9 billion. Consulting definitely scales.
Accenture is #493 of the S&P500, at $101k of revenue per employee. (!) They're doing a bit better than McDonalds (#497). The scalability argument holds -- compare to Netflix, which scales quite well (#23, $1.9M/employee).
I've noticed this pattern in both business and in individual engineers.
You either have a spectacular product that theoretically could sell itself with initial exposure, or you have the best marketing and a product that might as well be vapor.
Usually it's the mediocre product that's somewhere in between that wins.
In engineers I've noticed that some of the brightest I've worked with are unsociable, or have no self esteem, and get stuck in the same place for 20 years, and on the other side of the spectrum are talentless hacks that are great at selling themselves and moving around enough to increase their salary.
Meanwhile I feel like a fraud because I'm not the best, but good enough to recognize my worth and sociable enough to put myself out there.
I leverage the shit out of that and feel somewhat guilty about hoping that the really talented local engineers don't catch on to the idea.
Younger company with unknown decent product? More heavily spending on sales.
Older company with industry recommendations from previous clients? Spend less on fewer, more technical sales people.
Older company with no recommendations from previous clients? You're doing something wrong, and sales isn't the place you should be spending your money.
I feel like normally this logic is inverted at the C-level though. Sales are low and product isn't getting traction? Hire more sales and marketing folks and put off those engineering hires.
Consulting requires linear increase of resources. Software product requires logarithmic increase of resources. Therefore product business is more efficient to scale than consulting.
When people say "it scales" they are largely referring to the cost of serving incremental customers. Software does that well, consulting doesn't.