I'm fascinated by start-ups because of the sheer amount of energy on display within them.

By their nature, those who start businesses are optimistic souls, believing they have the solution, or at least a better one, to some problem. If they have sold others on that vision, founders can attract the talent and capital needed to turn those ideas into reality.

But few new firms are truly high-growth enterprises, which are as prone to frenetic expansions as they are to spectacular implosions. Plus, as exciting as it can be working for a start-up, there is usually an initial hit to income for those first employees.

No worries, right? After all, what about the potential for future riches from stock options or expected raises once sales begin to take off?

A recent report by a researcher at the University of Maryland and three others at the Census Bureau finds that real earnings at businesses with fewer than 20 employees and those less than two years old peaked in the early 2000s and have either fallen or stalled since.

Released by the Ewing Marion Kauffman Foundation, the report shows that workers at young firms earned about 85 percent as much as those at businesses 11 years old or older. That number had fallen to 70 percent by 2011.

Though it seems common sense that large companies pay better than small businesses, the gap between them appears to have grown. Average real monthly earnings at small firms decreased from 78 percent of that of established firms in 2001 to 66 percent in 2011.

In a series of reports, the entrepreneurship-focused Kauffman Foundation has been documenting the impact that start-ups and fast-growing businesses have on the U.S. economy.

This latest report begins to reveal a bit about the qualities of those jobs, and at least one conclusion was dispiriting: "Small employers, it seems, have become increasingly unable to match the wages offered by larger employers."

Why? The authors, who include Maryland economics professor John Haltiwanger, write that the answer will have to come from additional research.