Crunch optimizes the process of creating MapReduce jobs in Hadoop through Lazy Evaluation. It delays the execution of operations until the results are actually needed, reducing unnecessary computations and improving overall performance.

For advanced data analytics, Hadoop Streaming API can be coupled with Apache Pig to handle complex queries and computations. Pig provides a high-level scripting language, Pig Latin, making it easier to express data transformations and analytics tasks.

Snappy compression in Parquet allows for efficient storage and faster query processing. Snappy is a fast and lightweight compression algorithm, making it suitable for use in Big Data processing environments like Hadoop.

Advanced Hadoop applications might use CombineFileInputFormat for custom data processing requirements. This InputFormat combines small files into larger input splits, reducing the number of input splits and improving the efficiency of processing small files in Hadoop.

Scalability of a Hadoop cluster refers to its ability to handle the expected volume of data storage. A scalable cluster can easily accommodate growing data without compromising performance, making it a crucial aspect of Hadoop infrastructure design.

The number of DataNodes in a Hadoop cluster is crucially influenced by storage capacity. It determines how much data can be stored and processed concurrently across the cluster. Ensuring sufficient storage capacity is essential for optimal performance and data processing capabilities.

The JobTracker is an essential component in Hadoop for tracking job processing and resource utilization. It manages and schedules MapReduce jobs, tracks the progress of tasks, and monitors resource usage in the cluster. It plays a crucial role in coordinating job execution across the nodes.

HBase's in-memory storage feature makes it suitable for real-time read/write access. The data is stored in memory, enabling faster access for read and write operations, making it well-suited for applications requiring low-latency responses.

Reflection is a key feature in Avro that facilitates data serialization and deserialization in a distributed environment. It enables automatic generation of code for serialization and deserialization, simplifying the process of working with complex data structures.

In the Hadoop Streaming API, custom scripts are often used to optimize the mapping and reducing processes. These scripts, usually written in languages like Python or Perl, allow users to define their own logic for data transformation, filtering, and aggregation, providing flexibility and customization in Hadoop data processing.