A scalable feedback mechanism to solicit feedback from a potentially very large group of networked nodes is an important building block for many network protocols. Multicast transport protocols use it for negative acknowledgements and for delay and packet loss determination. Grid computing and peer-to-peer applications can use similar approaches to find nodes that are, at a given moment in time, best suited to serve a request. In sensor networks, such mechanisms allow to report extreme values in a resource efficient way. In this paper we analyze several extensions to the exponential feedback algorithm [5,6] that provide an optimal way to collect extreme values from a potentially very large group of networked nodes. In contrast to prior work, we focus on how knowledge about the value distribution in the group can be used to optimize the feedback process. We describe the trade-offs that have to be decided upon when using these extensions and provide additional insight into their performance by means of simulation. Furthermore, we briefly illustrate how sample applications can benefit from the proposed mechanisms.