A few thoughts:
Time-multiplex the data. You can cycle through the sets of statistics, displaying each set for about 10 seconds. That way, you can use the entire screen to show only one set. This is only good if it’s not particularly important for users to see every statistic all the time (e.g., the TV is intended to provide an occasional motivation boost or to give the workplace a high-tech real-time emotional feel). It’s also only good if the TV is placed such that the constantly changing screen doesn’t annoy or distract users from their work, but is nonetheless noticeable (e.g., it’s in the elevator lobby, where the users have little else to look at anyway).
Multiple TVs. You can show different sets of statistics on different screens. This is especially desirable if different users are interested in different statistics or time periods. You can arrange the TVs in one full-wall matrix, or (better) you can distribute the different TVs around the workplace so each group of users can most easily see the data it uses the most (assuming the users themselves are located around the workplace by function at more-or-less fixed workstations). This, of course, means greater hardware expense, and wall space consumption, which may not be available.
Eliminate low-priority data. You can review your user research and eliminate the statistics that are least useful for the purpose the TV serves. For example, for at-a-glance look, maybe the monthly data are not useful to see all day long, day after day, because they change so slowly. Let users access those data a different way (e.g., by subscribing to a monthly tweet). On the other hand, maybe the daily data are not useful without more context because they’re subject to so much random variation. Maybe those are better left to an interactive desktop app. Maybe weekly data say pretty much the same thing that monthly data say (they’re only separated by a factor of four), so one can be eliminated from the TV.
Create a combined metric. You can combine several related statistics into a single metric of performance. For example, you can take the weighted average of sign-ups, conversions, and downloads to represent the general activity level. Review your user research to see how users mentally combine statistics to get an idea of what would make useful combined metrics. This, of course, can mean training and educating users on a new metric.
Reduce display redundancy. Rather than label each statistic like you’ve done in your mockup (e.g., “Signups today”), you can provide row and column headers (e.g., Today, This Week, This Month across the top, and each statistic name down the left margin). This means fewer characters to fit in the same space so each character can be larger. If the statistics widgets show information related to the charts, integrate the statistics with the charts, reusing as many elements as possible (e.g., statistic names).
Consider graphic representation. Rather than displaying a grid of 48 individual numbers, you can combine related numbers into a single graphic. For example, you could show each statistic as a line graphic of the daily quantity over two months. You can superimpose on this graph the weekly and/or monthly moving average. This would show the daily and weekly/monthly relative levels and trends more clearly and precisely than six individual numbers (even with your color coding). Because the graphs will be small, with the y-axis values hard to read, this only works if users are primarily interested in relative levels and trends, not absolute values. That is, a line graph is better for seeing how much higher Statistic A has gotten (e.g., versus Statistic B); the grid it better if users are looking for a single precise numeric value (e.g., 210). There may be helpful ways to combine multiple related statistics into a single graphic too. Other graphic representations depend on the details of how the data are used.