My 2 cents...
I think it is all about optimizing / training your strategy against different patterns. The more patterns, the better. However, like with everything in life, there is a trade-off. Long-term data (i.e. many years) will include patterns that may never be seen again -- hence, you'll be training your strategy with irrelevant data -- which will skew the settings and yield poorer stats when your strategy encounters more recent patterns. Also, when you use many years worth of data then there are many more bars to calculate and optimizations take longer -- hence, limiting your ability to test lots of different strategies.
I think it is more about the number of bars and using recent data. As an example, for H4 I'm using around 3000 bars. I think that works out to about 16 months. Actually -- it's 8 months each of recent EURUSD and USDCHF stitched together.
When I do a sanity check against MT4 or MT5 back testers using longer data horizons, the results are roughly consistent with FSBPro -- which means that optimizing / training with 3000 bars may be sufficient. So, I'm gaining confidence emphasizing recent data horizons when generating and optimizing / training.
In a round about way I guess I'm agreeing with Blaiserboy and footon. I don't think it is essential to use your Broker's data -- especially if you are trading longer time frames.